Cut-Away Views, Section Views, and Ghosted Views The popularity of cut-away and ghosted views is

demon-strated by the fact that they can be found in all books on technical or medical illustrations [GMS^∗02,Hod03]. An au-tomatic generation of cut-away and ghosted views for polyg-onal data was introduced by Feiner and Seligmann [FS92].

They propose a family of algorithms that automatically iden-tify potentially obscuring objects and display them in a ghosted or cut-away view. The proposed algorithms exploit z-buffer rendering, therefore they are suitable for real-time interaction achieved by hardware acceleration. Interactive semi-transparent views, section views, and cut-away views for polygonal data have been recently revised by Diepstraten et al. [DWE02,DWE03]. Semi-transparent views unveil in-teresting objects obscured by other context information by increasing the transparency of the context. Diepstraten et al.

propose to adhere to an effective set of rules for the auto-matic generation of the discussed illustrative techniques. For semi-transparent illustrative views the following three rules should be taken into consideration:

• faces of transparent objects never shine through

• objects occluded by two transparent objects do not shine through

• transparency falls-off close to the edges of transparent ob-jects

For section views and cut-away views they propose to keep in mind seven other rules:

• inside and outside objects have to be distinguished from each other

• a section view is represented by the intersection of two half spaces

• the cut-out of a section view is aligned to the main axis of the outside object

• an optional jittering mechanism is useful for cut-outs

• a mechanism to make the walls visible is needed

• cut-outs consist of a single hole in the outside object

• interior objects should be visible from any given viewing angle

The mentioned algorithms and rules for cut-away views, section views, and ghosted views have been applied to polygonal data and are generally applicable in computer graphics. For an arbitrary clipping of volumetric data Weiskopf et al. [WEE03] propose a number of effective techniques to increase performance and visual quality. The implementation of clipping operations is mapped to com-modity graphics hardware to achieve interactive framerates.

Additionally to clipping all rendering computations are per-formed on the graphics hardware. Per-fragment operations estimate on-the-fly the visibility according to the clipping geometry and adjust the shading in areas where clipping oc-curs. In the following Sections2.1and 2.2we focus more on visualization related tasks. First we will discuss an ap-proach for automatic cut-away and ghosted views out of sci-entific volumetric data [VKG04]. This technique employs

Viola and Gröller / Smart Visibility in Visualization

Figure 3:Stages in the pipeline of importance-driven volume rendering: Volumetric features are assigned importance values (left image). The volume is traversed (center) in the importance compositing stage to estimate levels of sparseness (right). These are used to enhance or suppress particular parts of the volume. The resulting images then emphasize important features.

additional information about importance of a particular fea-ture. Then we will show the potential of such expressive views on a set of applications.

2.1. Importance-Driven Feature Enhancement

Traditionally features within the volume dataset are clas-sified by optical properties like color and opacity. With importance-driven feature enhancement we additionally as-sign one more dimension to features, which describes their importance. Importance encodes which features are the most interesting ones and have the highest priority to be clearly visible. Prior to the final image synthesis, the visibility of im-portant features is estimated. If less imim-portant objects are oc-cluding features that are more interesting, the less important ones are rendered more sparsely, e.g., more transparently.

If the same object does not cause any unwanted occlusions in other regions of the image, it is rendered more densely, e.g., opaque, in order to see its features more clearly. This allows to see all interesting structures irrespective if they are occluded or not, and the less important parts are still vis-ible as much as possvis-ible. Instead of using constant optical characteristics, which are independent from the viewpoint, we use several levels of sparseness for each feature. Levels of sparseness correspond to levels of abstraction, i.e., we do not assign a single optical characteristic, but several charac-teristics with smooth transitions in between. These multiple levels of sparseness allow the object to continuously change its visual appearance from a very dense representation to a very sparse one. Which level of sparseness will be chosen, is dependent on the importance of the particular object and the importance of objects in front and behind. The level of sparseness thus may continuously vary within a single fea-ture. For different viewpoints the same part of a feature may be represented with different levels of sparseness. To deter-mine the sparseness level for each object or parts thereof the rendering pipeline requires an additional step, which we call importance compositing. This step evaluates the occlu-sion according to the viewpoint settings, takes the impor-tance factor of each feature into account and assigns to each feature a particular level of sparseness. The final synthesis results in images with maximal visual information with

re-spect to the predefined object importance. The interrelation-ship between object importance, importance compositing, and levels of sparseness is depicted in Figure 3. The im-portance compositing traverses the whole volume to iden-tify object occlusions and assigns the corresponding level of sparseness to each object. Object importance translates to object visibility in the result image. This causes different rendering settings for the context object (with importance 0.1) in the area of the image which is covered by the focus object (importance0.7).

Figure4shows cut-away view of multi-dimensional volu-metric data of hurricane Isabel using importance-driven fea-ture enhancement. The important feafea-ture was hurricane eye selected by cylindrical proxy geometry. Inside the cylinder the total precipitation mixing ratio is visualized. Thanks to the cut-away view it is possible to have a clear view at this property close to the eye of the hurricane. Outside the cylin-der is the context area where the total cloud moisture is vi-sualized.

Figure 4:Cut-away visualization of multidimensional volu-metric data of hurricane Isabel.

Figure5illustrates ghosted view of scalar volumetric data of a Leopard gecko. The small internal structure (in yellow) of Leopard gecko dataset is the most interesting information and has been pre-segmented. Body is considered as context information. In the area of occlusion the visual representa-tion of gecko body is reduced to contours to have a clear view on the interesting internal organ.

2.2. Applications in Visualization

Expressive visualizations inspired by illustration tech-niques are useful for various visualization tasks. Straka et

c

The Eurographics Association 2005.

Viola and Gröller / Smart Visibility in Visualization

Figure 5:Ghosted visualzation using contours of CT scan of Leopard gecko.

al. [SvC^∗04] are applying a cut-away technique for CT-angiography of peripheral arteries in human legs. The goal is to have a clear view on the vessels, which are partially seg-mented by their centerline. For a clear understanding of the spatial arrangement it is necessary to visualize also bones and skin contours. To have an unobstructed view on the ves-sel for each viewpoint it is necessary to perform a cut in the bone. To avoid potential misinterpretations, the cut is clearly depicted as an artificial and sharp change in the data. This is illustrated in Figure1(d).

An extension to direct volume rendering that focuses on increasing the visibility of features has been proposed by Bruckner et al. [BGKG05]. This technique is known as illus-trative context-preserving volume rendering. The approach maps transparency to the strength of specular highlights.

This allows to seeinsidethe volume in the areas of high-lights. The human perception can easily complete the shape of partially transparent parts and therefore additional infor-mation can be shown here. A further parameter tunes the ratio between specularity and transparency. A depth param-eter dparam-etermines how far one can look inside a volumetric object (fuzzy clipping). Certain data value ranges can be ex-cluded from the transparency modulation to allow a clear view on specific (inner) structures. An example of illustra-tive context-preserving volume rendering is shown in Fig-ure6.

An interactive tool for cut-away and ghosting vi-sualizations has been recently proposed by Bruck-ner [BG05,BVG05]. The tool is denoted as VolumeShop and it is an interactive system which features advanced ma-nipulation techniques and illustrative rendering techniques to generate interactive illustrations directly from the volu-metric data. The system is using latest-generation texture-mapping hardware to perform interactive rendering using various kinds of rendering styles. It implements a multi-volume concept to enable individual manipulations of each volume part. The segmentation of the volumetric objects can be done directly via 3D painting. Apart from importance-driven visualization resulting into cut-away and ghosted views, VolumeShop features a label management to

intro-Figure 6:Illustrative context-preserving volume rendering showing interior structures of a human hand [BGKG05].

duce basic descriptions for the visualized data. To focus at a particular feature, this feature can be moved from its original spatial position. To indicate its original spatial position it is possible to display aghostthere, or add additional markers such as fanning or arrows.

Some ghosted visualizations generated using Vol-umeShop are shown in Figure7.

Figure 7:Interactive ghosted visualizations of the engine block and human head datasets [BG05,BVG05].

Previous applications of cut-away views are viewpoint-dependent, i.e., the shape and location of the cut is directly dependent on the viewpoint information. Volume cutting is another medical visualization technique that is related to cut-away views, but the cut shape is not influenced by view-point settings. Pflesser et al. [PPT^∗02] present an

interac-Viola and Gröller / Smart Visibility in Visualization tive drill-like tool for surgical training, which is based on the

multi-volume concept. Owada et al. [ONOI04] extend vol-ume cutting by incorporating two-dimensional textures that are mapped on the cut surface. This enhances the visualiza-tion with addivisualiza-tional informavisualiza-tion of the internal arrangement of bones or muscles. Such a concept can be very useful for anatomy education for example. Both volume cutting tech-niques are illustrated in Figure8.

Figure 8:Volume cutting featuring two-dimensional textures for anatomy education [ONOI04] (left) and volume cutting with drill-like tool for surgical education [PPT^∗02] (right).

Visualization of complex dynamical systems can be also enhanced by incorporating cuts into stream surfaces. Strea-marrows proposed by Löffelmann et al. [LMG97] exploit cutting for enhancing the visual information. They use ar-rows as a basic element for cutting away part of the stream surface. This allows to see through the surface and perceive other surfaces or structures behind. Animating streamarrows along the stream surface enables to see beyond the front stream surfaces and perceive the flow direction. Streamar-rows belong to the category of view-point independent cut-away techniques and are shown in Figure1(a).